Image Fusion Based On Optimization Technique

Mahendra kumar1 Shiven Goyal2

University Department of Electronics Engineering, University Department of Electronics Engineering,
Rajasthan Technical University, Rajasthan Technical University,
Kota (Rajasthan), India Kota (Rajasthan), India
Email: miresearchlab@gmail.com Email: shivenshines@gmail.com

Abstract In this paper a saras type-image-fusion technique hybrid curve-let and SOMA algorithm which provide fused
using hybrid Curve let and SOMA is used. The method shows image with superior quality.
fusion of images and can be used for fusion of multi model image.
It is concluded that fusion with advanced single levels offers II. OPTIMAL CURVE-LET TRANSFORM
better fusion quality. This method provides a comparative study
between proposed & literature techniques and validation of the Curve let transform is potent at spotting image activity along
projected algorithm as Peak Signal to Noise Ratio (PSNR), Root curves. Discrete version executed in this work uses Unequally
Mean Square Error (RMSE). Apart from the hybrid technique Spaced Fast Fourier Transform algorithm [USFFT]. The fast
used, it is founded that the performance parameters PSNR, discrete curve let transform is easier, rapid and less redundant
RMSE excelled as compared with the previous findings. [8, 9, 10, 11, 12,18,20]. This principle uses a decimated
rectangular grid tilted along the key direction of each curve let.
Keywords Saras type image, Curve let transform, PSNR, The windows Uj,i smoothly localizes Fourier transform
RMSE, SOMA near the cropped wedges obeying the scaling. Two parameters
are involved in the digital and number of angles at the coarsest
level. Since resolution extraction is two levels of resolutions.
I. INTRODUCTION
Multiple sensors data fusion provide an imperative role in
defense, satellite images and wide civilian applications as Input Image
eclectic mix of sensors are offered, working in diverse spectral
bands. A far-flung tool in signal and image processing is multi
scale disintegration. Augmentation of information content of
image is vital in biomedical science and multiple registered
images are annexed to get an image fusion but an afoot Apply 2D-FFT
research part. Various algorithms of image fusion like multiple
resolution [1,2], multiscale[3] and statistical signal processing
[4,5,6] based methods are presented and estimated.
Advancement in arena of sensing technologies, multi sensor Scale/angle pair(j,i) resample
techniques have become a literalism in variegated fields like
machine vision, medical imaging, remote sensing and military
operations for which they were elaborated.
Image fusion presents a potent means of decreasing Perform multiplication of interpolated object with Ui
the growing amount of data, concurrently extracting every
valuable data from rootb image. Increment in total amount of
data accessible is offered by these techniques. To craft new
images so as to get germane for the function of machine/human
perception, for additional image processing assignments like
medical imaging and remote sensing, also its applications like Obtain f[n1,n2]=f[n1,n2-n1tan l]Ui[n1,n2]
segmentation, target detection or object exposure are other
aims of image fusion. Multi sensor data frequently offers
corresponding data, so image fusion imparts an effectual
method for analysis of data and to enable comparison. As an Do inversion of 2D FET to every fi,l
example, visible band and infrared pictures can be fused to help
pilots landing aircrafts in deprived visibility [3,5,7]. At the
end, the functioning of the image fusion schemes is predicted Curve let Coefficient
as tradeoffs between true image & fused image. In preceding
techniques when apply saras type images, the performance
standard is deprived. As a result this paper proposed a novel Fig.1. CLT Algorithm
 2016 IEEE International Conference on Computational Intelligence and Computing Research
III. SELF ORGANIZING MIGRATING ALGORITHM
Self-Organizing Migration Algorithm is based on self-
arrangement behavior of batch of individuals in social
ambient. This can also be assorted as evolutionary algorithm
regardless of fact that no new productions of individuals are
yielded during hunt process (ideology of this algorithm). Only
places of individuals in the quested space are modified during
generation called migration loop. The algorithm, developed
by prof. Zelinka in 1999. Several different versions of SOMA
exist. This section presents the most common All-to-One
SOMA [19].
SOMA
Parameter defined: Step, Path length, Pop size, PRT and
Cost Function
Population of individuals is generated randomly.
Migration Loop: Pop size is evaluated by the cost
function and the leader (individuals with the highest
fitness) is selected for the current.
All other individuals start to jump, towards the leader.
Jumping continues until a new place defined by Path
Length.
The individuals then return to the position where it founds
the best fitness on its trajectory.
xiMLnew
,j = xiML ML ML
, jstart + ( x L , j x i , jstart )tPRTVector
Before an individual start jumping towards the leader, a
random number rnd is generated and compared with PRT.
If rndj<PRT then PRTVectorj=0 else 1, where rnd,
<0,1 >and j=1,nparam
Optimum value
Fig 2: SOMA Algorithm
IV. PROPOSED IMAGE FUSION SYSTEM
Consider image 1 as saras type image 1 and image 2 as
saras type image 2 and image compression and feature
extracted by curve let transform technique and apply Image
fusion algorithm as averaging method for fused image and
compare actual/true image with fused image and calculate
PSNR and RMSE to check effectiveness of proposed system
[4, 14, 15,16,17,18,20].
Pseudo code:
Im1=curvelet (image1)
Im1=I curvelet (De1)
Id1=Im1-Im1
Im2=curvelet (image2)
Im2=Icurvelet (De2)
Id2=Im2-Im2
Id=abs(Id1)-(Idf)>=0;
Imf=Idf+Imf
Imf=Icurvelet (Idfm)
This pseudo code shows the implementation Process [4, 14,
15, 16, 17].
Multiresolution methodology is profoundly related to image
operations, biological vision, computer visions, and scientific
computing. The curve let transform is a multiscale directional
transform that permits an almost optimal unadaptable sparse
representation of objects with edges. It has increased interest in
the community of applied mathematics and signal processing
over the years. Edges plays an elementary function in image
comprehensions, one good way to improve spatial resolution is
to improve the edges. Curve let based image fusion
methodology provides significant amount of information in the
spatial and spectral domains simultaneously. [18].
 2016 IEEE International Conference on Computational Intelligence and Computing Research

This value will be more when the fused and true images are
SOMA SOMA
selfsame and higher value indicates excelled fusion.

PSNR =

CURVE LET CURVE LET

TRANSFORM TRANSFORM 2
L
20 log 10 - (2)
1 M N

MN

X =1

Y =1
( I Y ( x , y ) I f ( x , y )) 2


FUSION
ALGORITHM Where, L is the number of gray levels in the image.

VI. RESULT AND COMPARATIVE STUDY

INVERSE CURVE LET TRANSFORM

True image img1.jpg of value 512x512 is revealed in figure-4.

FUSED IMAGE The two images to be mixed are produced from the true
images as revealed in figure 5&6. The fused image is nearly
Fig.3: Proposed Image Fusion Method analogous to true image and the faulty image is approximately
null. Table-1 reveals the fusion quality evaluation metrics. The
table shows proposed hybrid technique where metrics
demonstrated in dark font are superior amongst others. We got
V. FUSION PERFORMANCE EVALUATION
almost similar performance by using fusion with 3 level
pyramids or some higher values. By applying projected
The performance of image fusion algorithm can be valuated as algorithm results are improved as error values are reduced
following parameters [3, 5, 8, 9]: which enhances PSNR values.

Root Mean Square Error:

It is measured as the root mean square error (RMSE) of the
corresponding pixels in the true image It and the fused image
If. It will be nearly zero when the true and fused images are
identical and it will increase when the dissimilarity increases.

RMSE =

M N
1
MN
( I
X =1 Y =1
Y ( x, y ) I f ( x, y )) 2 - (1)

Fig4: Reference saras type image

Peak Signal to Noise Ratio:

 2016 IEEE International Conference on Computational Intelligence and Computing Research
Fig 5: saras type image 1
Fig 6: saras type image 2
VII CONCLUSION
An innovative image fusion technique using curve let
transform and SOMA has been proposed and its performance is
evaluated. We realized better fusion quality by enhancing the
fusion levels till 3. This method can be efficiently taken in use
for fusion of images with multiple spectral images also. The
projected algorithm is easier and efficient to implement for real
time applications. This paper also provides a comparative
study between proposed & literature techniques and
validation of the projected algorithm as in table 1.
REFERENCES
[1] A Toet, A morphological pyramid image decomposition, Pattern
Recognition Lett 9(4), 255261 (1989).
[2] V.P.S Naidu and Raol, Pixel-Level Image Fusion using Wavelets and
Principal Component Analysis A Comparative Analysis Defense Science
Journal, Volume.58, Number.3, pp(338-352), May 2008.
[3] VPS Naidu, Discrete Cosine Transform-based Image Fusion, Special
Issue on Mobile Intelligent Autonomous System, Defense Science Journal,
Volume. 60, Number.1, pp(48-54), January 2010.
[4] M.Kumar et.al., Digital Image Watermarking using Fractional Fourier
transform via image compression, In IEEE International Conference on
Computational Intelligence and Computing Research 2013 (IEEE ICCIC
2013), 26-28 December,2013.
[5] VPS Naidu, Novel Image Fusion Technique using DCT based Laplacian
Pyramid, International Journal of Inventive Engineering and Sciences, ISSN:
23199598, Volume-1, Issue-2, January 2013.
[6] R.S. Blum, Robust image fusion using a statistical signal processing
approach, Image Fusion, 6, pp(119-128), 2005.
[7] Shutao Li, J. T Kwok and Y. Wang, Combination of images with diverse
focuses using the spatial frequency, Information fusion, 2(3), pp.167-176,
2001.
[8] Mahendra Kumar et. Al..Fuzzy type Image Fusion using SPIHT Image
compression techniques, International Journal of Engineering Research and
Applications, (Accepted) (Impact factor: 1.69)
[9] Mahendra Kumar et. al., Digital Image Watermarking using Fractional
Fourier Transform with Different Attack International Journal of Scientific
Engineering and Technology (IJSET), Volume No.3 Issue number.8, August
2014, pp : 1008-1011. (ISSN : 2277-1581)
[10] M.Kumar, Face Recognition using SOM Neural Network with Different
Facial Feature Extraction Technique International Journal of Computer
Applications, Volume 76, Number 3, August 2013, pp.7-11. ((Impact factor:
0.821))
[11] Mahendra Kumar, A Matlab Based High Speed Face Recognition
System Using SOM Neural Networks International Journal of Engineering
Research and Applications (IJERA), Vol. 3, Issue 4,July-Aug. 2013, pp.785-
790. (Impact factor: 1.69)
[12] S. Prabha, Dr. M. Sasikala, "Texture Classification Using Curvelet
Transform", International Journal of Advancements in Research &
Technology(IJART), Vol. 2, Issue4, Apr. 2013.
[13] Rajesh Kumar Kakerda et. Al.,Fuzzy type Image Fusion using hybrid
DCT-FFT based Laplacian Pyramid Transform, 4th IEEE International
Conference on Communication and Signal Processing (ICCSP 2015) 02-04
April 2015 - Melmaruvathur, TN, IND.
[14] M. Kumar et. Al., Comparative Study Of Different Classifiers Based
Speaker Recognition System Using Modified MFCC For Noisy Environment,
International Conference Green Computing and Internet of Things (ICGCIoT -
2015) 08-10 Oct., 2015, Delhi, IND.
[15] M. Kumar et. Al., Robust Digital Image Watermarking using DCT based
Pyramid Transform via image compression, 4th IEEE International
Conference on Communication and Signal Processing (ICCSP 2015) 02-04
April 2015 - Melmaruvathur, TN, IND . (Published in IEEExplore)
[16] Mahendra Kumar et. Al.,Fuzzy type Image Fusion using hybrid DCT-
FFT based Laplacian Pyramid Transform, 4th IEEE International Conference
on Communication and Signal Processing (ICCSP 2015) 02-04 April 2015 -
Melmaruvathur, TN, IND.. (Published in IEEExplore)
[17] Mahendra Kumar et. Al., Digital Image Watermarking using Hybrid
DWT - FFT Technique with Different Attacks, 4th IEEE International
Conference on Communication and Signal Processing (ICCSP 2015) 02-04
April 2015 - Melmaruvathur, TN, IND. (Published in IEEExplore)
[18] Mahendra Kumar et. Al., Robust Image Fusion based on Optimal Cuve-
let Transform, International Conference on Micro-Electronics and
Telecommunication Engineering (ICMETE 2016) 22 & 23, September, 2016,
SRM University, Modinagar, Gaziabad, India.
[19] Mahendra Kumar et. Al., Image Fusion Based On Hybrid SPIHT and
SOMA, International Conference on Micro-Electronics and
Telecommunication Engineering (ICMETE 2016) 22 & 23, September, 2016,
SRM University, Modinagar, Gaziabad, India.
[20] Mahendra Kumar et. al., Image Fusion based on the Modified Curvelet
Transform In International Conference on Smart Trends for Information
Technology and Computer Communications (SmartCom 2016), Jaipur, India,
6 7 August, 2016. (Proceeding in CCIS Series of Springer ) (ISSN Number -
1865-0929).
[21] M.Kumar et. Al., Optimal Image Fusion, International Journal of
Computer Application (IJCA) (Accepted) for October 2016.
 2016 IEEE International Conference on Computational Intelligence and Computing Research

Table 1
Comparative Analysis

*Root Mean Square Error(RMSE);Percentage Fit Error(PFE);Mean Absolute Error(MAE);Correlation(CORR);Signal to Noise Ratio(SNR);Peak Signal to
Noise Ratio(PSNR);Mutual Information(MI);Universal Quality Index(UQI);Measure of Structural Similarity Index(SSIM)[3]

*Frequency Partition DCT Image fusion;Laplacian Pyramid DCT Image fusion(LPDCT);1D Multi Resolution DCT (MRDCT) [3]

Performance Parameter

Technique RMSE PFE MAE CORR SNR PSNR MI UQI SSIM

Proposed 0.0013 0.1443 0.00099068 1.0000 56.8149 77.0695 1.9924 1.0000 1.0000

SPIHT+SOMA[21] 0.0643 7.2180 0.0405 0.9974 22.8317 60.0814 1.5988 0.1255 0.9997

FPDCT[3] 18.2385 7.8243 16.8633 0.9971 22.1311 35.5549 1.1064 0.1521 0.7888

LPDCT[3] 3.4201 1.4672 1.6386 0.9999 36.6700 42.8244 1.2884 0.3728 0.9827

MRDCT[3] 7.1313 3.0593 4.9386 0.9995 30.2875 39.6331 1.1106 0.1568 0.8236